How AI breakthroughs push past the 3 bottlenecks in lifecycle marketing
In some sense, retention marketing should be easy – you have the customer, you know how to contact them, you know what they buy, when they buy it, and what comms they will and won’t engage with. All a marketer needs to do is send each customer the perfect message, with the perfect offer, through the perfect channel, with the optimal frequency to maximize revenue, purchases, or whatever metric matters most to the business. So why does this story sound more like a fairy tale?
In practice, there are three barriers to a lifecycle marketer’s dream approaching reality.
The data integration bottleneck. In theory, a business’ rich first-party data powers lifecycle marketing. In practice, data might be incomplete, jumbled, scattered, or disconnected from the platforms marketers use to orchestrate campaigns and communicate with customers.
The content creation bottleneck. To leverage the value of their customer data, marketers need to craft messages, subject lines, offers, and creative that will resonate with each individual. But making lots of variants takes time!
The experimentation bottleneck. Even if marketers have content, and have data to suggest which content and offers would best resonate with each customer, they still need to test and learn what works in practice. Traditionally marketers have tested their ideas using A/B testing, but A/B testing is slow and doesn’t scale.
But AI breakthroughs are changing the game. Generative AI built on large language models (LLMs), such as Google’s Bard and OpenAI’s ChatGPT, are poised to solve not only the content creation bottleneck, but the data integration bottleneck as well. However, generative AI is unlikely to be much help with the experimentation bottleneck – marketers will still need to test and learn what works with actual customers. To push through this bottleneck, savvy marketers are relying on a different kind of AI called reinforcement learning to automate the process of experimentation.
The data integration bottleneck
The martech ecosystem has seen a renaissance of data capabilities in the last 5 to 10 years, with the rise of data warehouses, customer data platforms (CDPs), and newer tools such as composable CDPs. Yet these tools are not without their challenges.
Data is often scattered, and different systems don’t talk to each other. Marketers typically plan and execute campaigns in marketing automation or orchestration platforms, which are not always connected to data systems in a way that delivers the insights marketers need to make decisions. An email sending platform will tell you about clicks and opens – though with new privacy restrictions even that data is of limited value – but deeper insights about customer behavior are located somewhere else. Often when marketers hope to draw insights from data, they realize they need to get a ticket on some data engineering team’s backlog first.
Unifying data in a warehouse or CDP, and making that data available and usable to marketers, is always a work in progress. There will always be data assets that haven’t been unified or integrated across systems, or data that is missing, or unusable because of errors. Marketers never seem to have clean nor complete data. But generative AI is poised to make a fundamental shift to this landscape. LLMs will automate data transformation and maintenance – detecting errors, transforming data to make it intelligible across systems, and ultimately building the pipelines which integrate systems.
These capabilities are quickly being productized – Lume AI, launched in March 2023, automates data integration by using AI to build and maintain pipelines between systems. Numbers Station, a Series A startup, lets a marketer question a database in natural language. These kinds of capabilities will automate the process of integrating data and systems. The days when marketers wait on data engineers are numbered.
The content creation bottleneck
As marketers begin to mine their data on existing customers, they’ll naturally develop ideas about how to best market to each segment, or better yet each individual. But the more fine-grained their data, the more content variants marketers will need if they want to extract value from that data with 1:1 messaging.
Generative AI is changing this game. Most of us are already using ChatGPT to help us draft emails or messages, but AI tools can do more than simply create content. AI chatbots can create variants of existing content, using an existing template as a base. This capability helps marketers be experimenters – testing different subject lines (perhaps with and without emojis) or customer responses to different plans and promotions.
The experimentation bottleneck
As generative AI automates data integration and streamlines variant creation, marketers come to the next bottleneck: experimentation. No matter how brilliant offers, incentives, messages, or creative seem on paper, marketers won’t know what works for each individual customer without testing and learning. Different customers, moreover, are more likely to engage with different channels, at different times, and on different days of the week. Manually testing all these options through A/B tests is slow, and doesn’t scale with the number of variables marketers need to test.
Here generative AI is much less likely to provide a breakthrough. A model like ChatGPT can interpret customer data, and can compose messages and variants, but can’t predict the optimal decision for each individual customer. If anything, the power of generative AI will make the pain of the experimentation bottleneck more acute. Marketers will soon be leveraging the ability to rapidly create content at scale, but how will they make good use of that content without knowing which variants are working? Fortunately for marketers, another type of AI, called reinforcement learning, is perfectly suited to automating the process of experimentation.
Here’s how it works:
First, choose a success metric you want to maximize, such as revenue, conversions, ARPU, or any other KPI you can measure from your data.
Next, choose dimensions along which you’d like to test (e.g., offer, subject line, creative, channel, timing, cadence, etc.)
Finally, choose the options available for each dimension, perhaps including email or message variants developed using generative AI.
From there, a reinforcement learning AI automates the process of experimentation. A platform like OfferFit makes daily recommendations for each customer, seeking to maximize the chosen success metric. The AI learns from every customer interaction, and applies these insights to the next day's recommendations.
Lifecycle marketers are closer than they’ve ever been to the dream of 1:1 decisioning for every customer, thanks to the twin technologies of generative AI and reinforcement learning. But to leverage the transformative power of new technologies, marketers will need to break through all three bottlenecks: data integration, variant creation, and experimentation. Generative AI will help marketers integrate data and create variants, while reinforcement learning helps them automate experimentation.
Ready to make the leap from A/B to AI?